☀️Epic Sunset☀️ Pairing☀️ · Open Skill 公開スキル
☀️Epic Sunset☀️ Pairing☀️
☀️Epic Sunset☀️ Pairing☀️
Health
0 0 1
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4150474/2026-10-15/449ntqus.jpg
# ☀️Epic Sunset☀️ Pairing☀️
Repaint the user's uploaded reference images into a unified "epic sunset" visual finish. 1 input image = repaint the subject into a new setting; 2 input images = **pair** both subjects into a single sunset scene.
There is no structured project state; cross-turn reuse relies only on the asset URLs already on the canvas and the text setup in the conversation.
## When to use
- The user uploads 1–2 images and says "make it an epic sunset", "pair them", "sunset version", "swap in a sunset background".
- The user only wants to change one link in the chain: only the outfit, only the background, only the lighting, or only putting two people together.
- The user asks to keep the subject consistent (face, hairstyle, body type, props unchanged) and change only the environment and clothing.
## Visual spec (must be written into the prompt as a whole)
The whole frame serves the single theme of "epic sunset"; color, material and light must all be unified:
- **Light**: low sun behind the subject → rim light traces a golden edge along the subject's outline; volumetric light / Tyndall beams through the air; long shadows stretching toward the camera; slight bloom on the edges.
- **Color**: dominant tones are warm amber, gold, burnt orange; the sky grades from orange-gold into magenta and deep violet; shadows use a cooler violet-blue complementary tone for contrast; overall high contrast, warm cinematic film grade, **no cold white light, no grey studio backdrop, no noon overhead sun**.
- **Material**: warm translucency in fabric, warm highlights on metal, luminous warm skin tones; wind-lifted cloth and hair.
- **Atmosphere**: floating dust particles, a lowered cloud layer, one restrained lens flare, atmospheric perspective (distance haze).
- **Forbidden**: watermarks, text, borders, modern cold light sources that clash with the theme, fantasy-style repaints that change the original subject's identity.
## Steps
1. **Read the images and lock subject consistency**. For each user reference image call:
`recognize_image(image_url=<reference image>, prompt="Describe the main subject in detail: identity traits (face shape, hairstyle/hair color, age impression, body type), current clothing and its material/color, pose, composition, and notable props. Be factual.")`
Condense the result into a short subject description to use as `{subject}` in the prompts below. **Identity traits must be preserved as-is**; only clothing, environment, lighting and atmosphere may change.
2. **(Optional) Fetch additional reference material**. If the user explicitly asks for "a different style of outfit / a different kind of background", search the asset library first:
- `search_library(keyword="epic sunset sky", scope="all")` → `get_library_item(item_id=...)` to get `media[].uri`
- Same for clothing: `keyword="golden hour outfit"`, etc., using the user's own wording / category words.
- Use only the real `uri` values returned; **never invent** URLs. If nothing suitable exists, just describe it in text.
3. **Call generation** (prompts are always in English; reply to the user in the user's language):
- **One reference image → repaint the setting / change outfit**:
`generate_image(prompt=<template A below>, input_urls=[reference image], model='i2i.hd', orientation='portrait', name='<short title, e.g. Epic Sunset>', entity_type='characters')`
- **Two reference images → pair into one frame**:
`generate_image(prompt=<template B below>, input_urls=[reference image A, reference image B], model='i2i.hd', orientation='landscape', name='<both subject names> Sunset Pair', entity_type='characters')`
The order of `input_urls` = left-to-right order in the frame; the 1st is on the left.
- **No reference images at all (text only)**: use `model='t2i'` and leave `input_urls` empty.
Parameter conventions:
- `model` must be a **channel**, never a physical model name; use `i2i.hd` when higher quality is needed